| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package nvme |
| 4 | |
| 5 | import ( |
| 6 | "bytes" |
| 7 | "encoding/json" |
| 8 | "strconv" |
| 9 | "strings" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/netdata/netdata/go/plugins/logger" |
| 13 | "github.com/netdata/netdata/go/plugins/plugin/go.d/pkg/ndexec" |
| 14 | ) |
| 15 | |
| 16 | type nvmeDeviceList struct { |
| 17 | Devices []struct { |
| 18 | DevicePath string `json:"DevicePath"` |
| 19 | Firmware string `json:"Firmware"` |
| 20 | ModelNumber string `json:"ModelNumber"` |
| 21 | SerialNumber string `json:"SerialNumber"` |
| 22 | } |
| 23 | } |
| 24 | |
| 25 | func (n *nvmeDeviceList) UnmarshalJSON(b []byte) error { |
| 26 | type plain nvmeDeviceList |
| 27 | |
| 28 | if err := json.Unmarshal(b, (*plain)(n)); err != nil { |
| 29 | return err |
| 30 | } |
| 31 | |
| 32 | if len(n.Devices) > 0 && n.Devices[0].DevicePath != "" { |
| 33 | return nil |
| 34 | } |
| 35 | |
| 36 | n.Devices = n.Devices[:0] |
| 37 | |
| 38 | var v211Format struct { |
| 39 | Devices []struct { |
| 40 | Subsystems []struct { |
| 41 | Controllers []struct { |
| 42 | SerialNumber string `json:"SerialNumber"` |
| 43 | ModelNumber string `json:"ModelNumber"` |
| 44 | Firmware string `json:"Firmware"` |
| 45 | Namespaces []struct { |
| 46 | NameSpace string `json:"NameSpace"` |
| 47 | } `json:"Namespaces"` |
| 48 | } `json:"Controllers"` |
| 49 | } `json:"Subsystems"` |
| 50 | } `json:"Devices"` |
| 51 | } |
| 52 | |
| 53 | if err := json.Unmarshal(b, &v211Format); err != nil { |
| 54 | return err |
| 55 | } |
| 56 | |
| 57 | for _, device := range v211Format.Devices { |
| 58 | for _, subsystem := range device.Subsystems { |
| 59 | for _, controller := range subsystem.Controllers { |
| 60 | for _, namespace := range controller.Namespaces { |
| 61 | devPath := namespace.NameSpace |
| 62 | if !strings.HasPrefix(devPath, "/dev/") { |
| 63 | devPath = "/dev/" + devPath |
| 64 | } |
| 65 | n.Devices = append(n.Devices, struct { |
| 66 | DevicePath string `json:"DevicePath"` |
| 67 | Firmware string `json:"Firmware"` |
| 68 | ModelNumber string `json:"ModelNumber"` |
| 69 | SerialNumber string `json:"SerialNumber"` |
| 70 | }{ |
| 71 | DevicePath: devPath, |
| 72 | Firmware: controller.Firmware, |
| 73 | ModelNumber: controller.ModelNumber, |
| 74 | SerialNumber: controller.SerialNumber, |
| 75 | }) |
| 76 | } |
| 77 | } |
| 78 | } |
| 79 | } |
| 80 | |
| 81 | return nil |
| 82 | } |
| 83 | |
| 84 | // See "Health Information Log Page" in the Current Specification Version |
| 85 | // https://nvmexpress.org/developers/nvme-specification/ |
| 86 | type nvmeDeviceSmartLog struct { |
| 87 | CriticalWarningValue nvmeNumber |
| 88 | CriticalWarning json.RawMessage `json:"critical_warning"` |
| 89 | Temperature nvmeNumber `json:"temperature"` |
| 90 | AvailSpare nvmeNumber `json:"avail_spare"` |
| 91 | SpareThresh nvmeNumber `json:"spare_thresh"` |
| 92 | PercentUsed nvmeNumber `json:"percent_used"` |
| 93 | DataUnitsRead nvmeNumber `json:"data_units_read"` |
| 94 | DataUnitsWritten nvmeNumber `json:"data_units_written"` |
| 95 | HostReadCommands nvmeNumber `json:"host_read_commands"` |
| 96 | HostWriteCommands nvmeNumber `json:"host_write_commands"` |
| 97 | ControllerBusyTime nvmeNumber `json:"controller_busy_time"` |
| 98 | PowerCycles nvmeNumber `json:"power_cycles"` |
| 99 | PowerOnHours nvmeNumber `json:"power_on_hours"` |
| 100 | UnsafeShutdowns nvmeNumber `json:"unsafe_shutdowns"` |
| 101 | MediaErrors nvmeNumber `json:"media_errors"` |
| 102 | NumErrLogEntries nvmeNumber `json:"num_err_log_entries"` |
| 103 | WarningTempTime nvmeNumber `json:"warning_temp_time"` |
| 104 | CriticalCompTime nvmeNumber `json:"critical_comp_time"` |
| 105 | ThmTemp1TransCount nvmeNumber `json:"thm_temp1_trans_count"` |
| 106 | ThmTemp2TransCount nvmeNumber `json:"thm_temp2_trans_count"` |
| 107 | ThmTemp1TotalTime nvmeNumber `json:"thm_temp1_total_time"` |
| 108 | ThmTemp2TotalTime nvmeNumber `json:"thm_temp2_total_time"` |
| 109 | } |
| 110 | |
| 111 | func (n *nvmeDeviceSmartLog) UnmarshalJSON(b []byte) error { |
| 112 | type plain nvmeDeviceSmartLog |
| 113 | |
| 114 | if err := json.Unmarshal(b, (*plain)(n)); err != nil { |
| 115 | return err |
| 116 | } |
| 117 | |
| 118 | if n.CriticalWarning != nil { |
| 119 | var v211Format struct { |
| 120 | Value int `json:"value"` |
| 121 | } |
| 122 | if err := json.Unmarshal(n.CriticalWarning, &v211Format); err == nil { |
| 123 | n.CriticalWarningValue = nvmeNumber(strconv.Itoa(v211Format.Value)) |
| 124 | } else { |
| 125 | n.CriticalWarningValue = nvmeNumber(n.CriticalWarning) |
| 126 | } |
| 127 | } |
| 128 | |
| 129 | return nil |
| 130 | } |
| 131 | |
| 132 | // nvme-cli 2.1.1 exposes some values as strings |
| 133 | type nvmeNumber string |
| 134 | |
| 135 | func (n *nvmeNumber) UnmarshalJSON(b []byte) error { |
| 136 | *n = nvmeNumber(bytes.Trim(b, "\"")) |
| 137 | return nil |
| 138 | } |
| 139 | |
| 140 | type nvmeCli interface { |
| 141 | list() (*nvmeDeviceList, error) |
| 142 | smartLog(devicePath string) (*nvmeDeviceSmartLog, error) |
| 143 | } |
| 144 | |
| 145 | // ndsudoNvmeCliExec executes nvme via ndsudo (Linux/BSD) |
| 146 | type ndsudoNvmeCliExec struct { |
| 147 | timeout time.Duration |
| 148 | } |
| 149 | |
| 150 | func (n *ndsudoNvmeCliExec) list() (*nvmeDeviceList, error) { |
| 151 | bs, err := ndexec.RunNDSudo(nil, n.timeout, "nvme-list") |
| 152 | if err != nil { |
| 153 | return nil, err |
| 154 | } |
| 155 | |
| 156 | var v nvmeDeviceList |
| 157 | if err := json.Unmarshal(bs, &v); err != nil { |
| 158 | return nil, err |
| 159 | } |
| 160 | |
| 161 | return &v, nil |
| 162 | } |
| 163 | |
| 164 | func (n *ndsudoNvmeCliExec) smartLog(devicePath string) (*nvmeDeviceSmartLog, error) { |
| 165 | bs, err := ndexec.RunNDSudo(nil, n.timeout, "nvme-smart-log", "--device", devicePath) |
| 166 | if err != nil { |
| 167 | return nil, err |
| 168 | } |
| 169 | |
| 170 | var v nvmeDeviceSmartLog |
| 171 | if err := json.Unmarshal(bs, &v); err != nil { |
| 172 | return nil, err |
| 173 | } |
| 174 | |
| 175 | return &v, nil |
| 176 | } |
| 177 | |
| 178 | // directNvmeCliExec executes nvme directly (Windows) |
| 179 | type directNvmeCliExec struct { |
| 180 | *logger.Logger |
| 181 | |
| 182 | nvmePath string |
| 183 | timeout time.Duration |
| 184 | } |
| 185 | |
| 186 | func (n *directNvmeCliExec) list() (*nvmeDeviceList, error) { |
| 187 | bs, err := ndexec.RunDirect(n.Logger, n.timeout, n.nvmePath, "list", "--output-format=json") |
| 188 | if err != nil { |
| 189 | return nil, err |
| 190 | } |
| 191 | |
| 192 | var v nvmeDeviceList |
| 193 | if err := json.Unmarshal(bs, &v); err != nil { |
| 194 | return nil, err |
| 195 | } |
| 196 | |
| 197 | return &v, nil |
| 198 | } |
| 199 | |
| 200 | func (n *directNvmeCliExec) smartLog(devicePath string) (*nvmeDeviceSmartLog, error) { |
| 201 | bs, err := ndexec.RunDirect(n.Logger, n.timeout, n.nvmePath, "smart-log", devicePath, "--output-format=json") |
| 202 | if err != nil { |
| 203 | return nil, err |
| 204 | } |
| 205 | |
| 206 | var v nvmeDeviceSmartLog |
| 207 | if err := json.Unmarshal(bs, &v); err != nil { |
| 208 | return nil, err |
| 209 | } |
| 210 | |
| 211 | return &v, nil |
| 212 | } |